EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Drop 0.10.2 (#3284)

Browse Source 
Drop v0.10.2 support, we support vLLM 0.11.0rc3 now.
- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/releases/v0.11.0

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This commit is contained in:
wangxiyuan
2025-10-09 10:28:38 +08:00
committed by GitHub
parent 2dde1268c7
commit f12f76d7ba
17 changed files with 202 additions and 653 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
vllm_ascend/models/qwen2_5_vl_without_padding.py
View File

@@ -68,7 +68,6 @@ from vllm.model_executor.models.utils import WeightsMapper, maybe_prefix
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm_ascend.models.qwen2_5_vl import AscendQwen2_5_VisionRotaryEmbedding
from vllm_ascend.utils import vllm_version_is
class AscendQwen2_5_VisionAttention_Without_Padding(Qwen2_5_VisionAttention):
@@ -484,20 +483,12 @@ class AscendQwen2_5_VLForConditionalGeneration_Without_Padding(
super().__init__(vllm_config=vllm_config, prefix=prefix)
config: Qwen2_5_VLConfig = vllm_config.model_config.hf_config
quant_config = vllm_config.quant_config
if vllm_version_is("0.10.2"):
self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
vision_config=config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=self._maybe_ignore_quant_config(quant_config),
prefix=maybe_prefix(prefix, "visual"),
)
else:
self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
vision_config=config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
)
self.visual = AscendQwen2_5_VisionTransformer_Without_Padding(
vision_config=config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
)
def _process_image_input(self, image_input) -> tuple[torch.Tensor, ...]:
@@ -563,20 +554,12 @@ class AscendQwen3VLForConditionalGeneration(Qwen3VLForConditionalGeneration):
super().__init__(vllm_config=vllm_config, prefix=prefix)
config: Qwen3VLConfig = vllm_config.model_config.hf_config
quant_config = vllm_config.quant_config
if vllm_version_is("0.10.2"):
self.visual = AscendQwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=self._maybe_ignore_quant_config(quant_config),
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel)
else:
self.visual = AscendQwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel)
self.visual = AscendQwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel)
@MULTIMODAL_REGISTRY.register_processor(Qwen3VLMultiModalProcessor,
@@ -613,19 +596,10 @@ class AscendQwen3VLMoeForConditionalGeneration(
multimodal_config = vllm_config.model_config.multimodal_config
self.multimodal_config = multimodal_config
self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"
if vllm_version_is("0.10.2"):
self.visual = AscendQwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=self._maybe_ignore_quant_config(quant_config),
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)
else:
self.visual = AscendQwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)
self.visual = AscendQwen3_VisionTransformer(
config.vision_config,
norm_eps=getattr(config, "rms_norm_eps", 1e-6),
quant_config=quant_config,
prefix=maybe_prefix(prefix, "visual"),
use_data_parallel=self.use_data_parallel,
)